Google ships Gemini 3.8 Live: SOTA realtime audio model across 97 languages
minchoi · x · 2026-09-16
Google DeepMind's Logan Kilpatrick announced Gemini 3.8 Live and 3.8 Live Extended Thinking, calling them new SOTA live audio models at frontier price + performance. 3.8 Live supports 97 languages with seamless switching, async tool calls, and more; the Extended Thinking variant adds reasoning to realtime voice.
Related event: Google Launches Gemini 3.8 Live Models, Topping Realtime Voice Benchmarks(14 posts)→
More from Models
- AI2's NGU sampling fixes RL for LLMs that only improves easy tasks — allenai · 2026-09-16
- With retries and pooled selection, Qwen3.8 27B hits 92.04% on DeepSWE 1.1, ~18 pts above GPT-6 Astra — S_Conradi · 2026-09-16
- Rohan Paul: With MCP Behind Every Model, Picking One Becomes Optional — kevinkern · 2026-09-16
- Redditor Claims Cursor's Grok 4.6 Gave an Oddly Self-Aware Reply — ISmellARatt · 2026-09-16
- Jev Benchmark Launches: $42 per Billion Input Tokens, Output Free Forever — cephaloform · 2026-09-16
- Same Prompt, Four Models Behind One MCP: Only One Got It Right — rohanpaul_ai · 2026-09-16